Skip to contents

Fit a mixture of von Mises distributions

Source: R/vonmises-mixture.R
fit_vonmises_mixture.Rd

Fits a finite mixture of von Mises components using an expectation maximization algorithm. For axial data, the algorithm fits doubled angles and returns component means on the original modulo-pi scale.

Usage

fit_vonmises_mixture(
  x,
  k = 2,
  weights = NULL,
  axial = FALSE,
  init = c("kmeans", "spaced"),
  nstart = 1,
  init_params = NULL,
  kappa_max = 10000,
  min_component_weight = 1e-08,
  max_iter = 200,
  tol = 1e-08,
  na.rm = TRUE,
  seed = NULL
)

Arguments

x

Numeric vector of angles in radians.

k

Number of mixture components.

weights

Optional non-negative observation weights.

axial

Should data be treated as axial, modulo pi?

init

Initialization method, either "kmeans" or "spaced".

nstart

Number of EM starts. The fit with the largest log-likelihood is retained.

init_params

Optional list or data frame with initial proportions, mu and kappa values.

kappa_max

Maximum fitted concentration. This caps nearly degenerate components.

min_component_weight

Minimum relative component weight before a component is reinitialized.

max_iter

Maximum number of EM iterations.

tol

Convergence tolerance on the log-likelihood.

na.rm

Should missing values be removed?

seed

Optional random seed used for initialization.

Value

An object of class ggcircular_vonmises_mixture.

See also

Other circular distributions: stat_vonmises(), stat_vonmises_fit(), stat_vonmises_mixture()

Examples

fit <- fit_vonmises_mixture(wind_directions$direction, k = 2)
#> Warning: `fit_vonmises_mixture()` did not converge within `max_iter` iterations.
tidy_circular(fit)
#> # A tibble: 2 × 4
#>   component proportion    mu kappa
#>       <int>      <dbl> <dbl> <dbl>
#> 1         1      0.321 0.903 1.49 
#> 2         2      0.679 4.06  0.755